A Hybrid LSTM–Grey Wolf Optimization Framework for Optimal Siting and Sizing of Distributed Generation in Radial Distribution Networks: Evidence from IEEE Benchmark Systems and A Nigerian 33 kV Feeder

Abstract : Electric power distribution networks in developing countries, particularly Nigeria, continue to experience high technical losses, poor voltage profiles, and increasing load demand on ageing, radially configured feeders. Although distributed generation is a proven technical solution to these challenges, its benefits depend heavily on optimal siting and sizing, a problem that is non-linear, multi-objective, and difficult to solve using conventional methods. This study developed and validated a hybrid intelligent framework that combines a Long Short-Term Memory neural network for load forecasting with a Grey Wolf Optimizer for distributed generation placement. The framework was tested on the IEEE 33-bus and IEEE 69-bus benchmark systems and, uniquely, on one hundred and twenty-five days of real operational data collected from the Moniya-Akinyele 33 kilovolt distribution feeder in Ibadan, Nigeria. Load flow analysis using the Newton-Raphson method identified weak buses through a composite priority index combining voltage deviation and loss sensitivity. On the benchmark system, the proposed framework reduced real power losses by 75.3 percent and improved the minimum bus voltage from 0.6746 to 0.6954 per unit following distributed generation installation of 243.88 kilowatts across seven priority buses. On the Nigerian feeder, the forecasting model achieved a coefficient of determination of 0.9667 and a mean absolute percentage error of 3.83 percent using real operating data, substantially outperforming its performance on synthesised benchmark data. The findings demonstrate that combining machine learning-based demand forecasting with meta-heuristic optimisation provides a practical and technically sound approach for intelligent network planning in developing-country distribution systems.